Latent Perspective-Taking via a Schr\"odinger Bridge in Influence-Augmented Local Models

2026-02-02

AI Summary

This work proposes a scalable and generalizable social reasoning framework to enable robots to effectively infer others’ latent mental states in human-robot cohabitation environments. The approach integrates a neuro-symbolic world model to construct a factored discrete dynamic Bayesian network, decomposing social tasks into local dynamics, social influence, and exogenous factors. It further introduces an amortized perspective-taking operator grounded in the Schrödinger bridge formulation, enabling efficient transformation from egocentric to allocentric belief representations. Evaluated on MiniGrid social navigation tasks, the architecture successfully supports model-based reinforcement learning, yielding socially aware policies that exhibit both interpretability and strong generalization capabilities.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Knowledge Representation and Reasoning: Reasoning with Beliefs
• Intelligent Robots: State Estimation

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Social networks and social learning
• User Modeling, Personalization and Recommendation: Social recommender systems and personalization for the social Web

Abstract

Operating in environments alongside humans requires robots to make decisions under uncertainty. In addition to exogenous dynamics, they must reason over others'hidden mental-models and mental-states. While Interactive POMDPs and Bayesian Theory of Mind formulations are principled, exact nested-belief inference is intractable, and hand-specified models are brittle in open-world settings. We address both by learning structured mental-models and an estimator of others'mental-states. Building on the Influence-Based Abstraction, we instantiate an Influence-Augmented Local Model to decompose socially-aware robot tasks into local dynamics, social influences, and exogenous factors. We propose (a) a neuro-symbolic world model instantiating a factored, discrete Dynamic Bayesian Network, and (b) a perspective-shift operator modeled as an amortized Schr\"odinger Bridge over the learned local dynamics that transports factored egocentric beliefs into other-centric beliefs. We show that this architecture enables agents to synthesize socially-aware policies in model-based reinforcement learning, via decision-time mental-state planning (a Schr\"odinger Bridge in belief space), with preliminary results in a MiniGrid social navigation task.

Problem

• mental-model
• perspective-taking
• social reasoning
• belief inference
• human-robot interaction

Innovation

• Schrödinger Bridge
• Influence-Augmented Local Model
• Theory of Mind
• Neuro-Symbolic World Model
• Perspective-Taking